Bidirectional Long Short-Term Memory based Gated Convolutional Network with Attention Mechanism for Traffic Flow Prediction

The traffic flow prediction becomes significant role in assuring a safety as well as solving traffic congestions. The difficult spatial temporal relationships of traffic road capturing are significant and this is the greater challenge. The manual approaches significantly capture the spatial and temporal relationships as well as enhance the model complexity through examining both connected and unconnected roads. To address to challenge, this research develops a Bi-directional Long Short-Term Memory (BiLSTM) with Attention Mechanism for the efficient traffic flow prediction. This research utilized the two datasets named PEMS04 and PEMS08 for validating a performance of the developed method. In the pre-processing phase, a normalization technique is utilized for normalizing the traffic data. A GCN is utilized for extracting the features and BiLSTM is utilized for classification. The developed approach reaches superior results through considering large performance indices and it achieves the accuracy of 0.9467 and 0.9534 in PEMS04 and PEMS08 dataset when compared to existing methods like Local and Global spatial GCN (T-LGGCN) and Spatial-Temporal Attention Graph Convolutional Network on Edge Cloud approach (STAGCN-EC).